UY-NET: A Two-Stage Network to Improve the Result of Detection in Colonoscopy Images

Abstract

The human digestive system is susceptible to various viruses and bacteria, which can lead to the development of lesions, disorders, and even cancer. According to statistics, colorectal cancer has been a leading cause of death in Taiwan for years. To reduce its mortality rate, clinicians must detect and remove polyps during gastrointestinal (GI) tract examinations. Recently, colonoscopies have been conducted to examine patients’ colons. Even so, polyps sometimes remain undetected. To help medical professionals better identify abnormalities, advanced deep learning algorithms that can accurately detect colorectal polyps from images should be developed. Prompted by this proposition, the present study combined U-Net and YOLOv4 to create a two-stage network algorithm called UY-Net. This new algorithm was tested using colonoscopy images from the Kvasir-SEG dataset. Results showed that UY-Net was significantly accurate in detecting polyps. It also outperformed YOLOv4, YOLOv3-spp, Faster R-CNN, and RetinaNet by achieving higher spatial accuracy and overall accuracy of object detection. As the empirical evidence suggests, two-stage network algorithms like UY-Net will be a reliable and promising aid to image detection in healthcare.